Paul Pop, Petru Eles, Zebo Peng Embedded Systems Lab (ESLAB) Linköping University, Sweden

1. Schedulability Analysis and Optimization for the Synthesis of Multi-Cluster Distributed Embedded Systems Paul Pop, Petru Eles, Zebo PengEmbedded Systems Lab (ESLAB)Linköping University, Sweden

2. Distributed Heterogeneous System NoCs ... ... ... ... ... Factory Systems Heterogeneous NetworksMulti-Cluster Systems Automotive Electronics Heterogeneous Networks

3. ... • Applications distributed over the heterogeneous networks • Reduce costs:use resources efficiently • Requirements:close to sensors/actuators Gateway ... • Applications distributed over heterogeneous networks are difficult to... • Analyze(e.g., guaranteeing timing constraints) • Design (e.g., efficient implementation) Unsolvedproblems Distributed Safety-Critical Applications

4. Contributions • Design optimization • Several design problems can be addressed once an analysis is available. • Analysis and design of Multi-Cluster Embedded Systems • Analysis and design of Multi-Cluster Embedded Systems • Analysis • Proposed a schedulability analysis for safety-critical hard real-time applications mapped on multi-cluster distributed embedded systems • Is the application schedulable? (Are deadlines satisfied?) • Bounds on the communication delays and communication buffer sizes • Design optimization • In this paper we have addressed communication synthesis and priority assignment for • Improving the degree of schedulability of an application • Minimizing communication buffer sizes needed to run a schedulable application

5. Outline • Motivation • Contributions • System architecture and application model • Schedulability analysis for multi-clusters • Optimization strategies • Experimental results • Message and future work

6. ... Time-triggered cluster • Static cyclic scheduling • Time-triggered protocol Gateway ... Event-triggered cluster • Fixed priority preemptive scheduling • Controller area network protocol Time Triggered Protocol (TTP) • Bus access scheme: time-division multiple-access (TDMA) • Schedule table located in each TTP controller: message descriptor list (MEDL) Controller Area Network (CAN) • Priority bus, collision avoidance • Highest priority message wins the contention • Priorities encoded in the frame identifier S0 S1 S2 SG S0 S1 S2 SG Slot TDMA Round Cycle of two rounds Hardware Architecture

7. ... ... m3 m1 m1 m2 m3 Software Architecture Gateway Time-triggered cluster NG N2 N1 CPU CPU CPU CPU CAN Controller CAN Controller P1 P4 OutCAN OutTTP OutN2 T T P2 P3 MBI MBI TTP Controller TTP Controller Event-triggered cluster SG S1 SG S1 Round2

8. ... ... • Output • Worst case response times and bounds on the buffer sizes • Design implementation such that the application is schedulable and buffer sizes are minimized • Schedule table for TT processes • Priorities for ET processes • Schedule table for TT messages • Priorities for ET messages • TT bus configuration (TDMA slot sequence and sizes) System configurationparameters Application: set of process graphs Mapping Architecture: Multi-cluster Communicationinfrastructureparameters Problem Formulation • Input • An application modeled as a set of process graphs • Each process has an worst case execution time, a period, and a deadline • Each message has a known size • The system architecture and the mapping of the application are given

9. Response times depend on the communication delay between sending and receiving a message • Communication delays depend on the type of message passing • TTCTTC • TTCETC • ETCETC • ETCTTC • Communication delays • Bounds on the buffer sizes Schedulability Analysis • Scheduling time-triggered activities: • Building a schedule table: static cyclic scheduling (e.g., list scheduling) • Scheduling event-triggered activities: • Response time analysis: calculate worst case response times for each process • Schedulability test: response times smaller than the deadlines

10. Multi-Cluster Scheduling • Scheduling cannot be addressed separately for each type of cluster • The inter-cluster communication creates a circular dependency: • TTC static schedules (offsets)  ETC response times • ETC response times  TTC schedule table construction Application, Mapping,Architecture TT Bus Configuration Priorities Offsetearliest possible start time for an event-triggered activity ResponseTimes ResponseTimeAnalysis StaticScheduling Offsets Multi-Cluster Scheduling ScheduleTables Responsetimes Boundson thebuffer sizes

11. ... ... P4 P4 P4 P1 P1 P1 N1 N1 N1 S1 m1 m2 m1 m1 m2 m2 m3 m3 m3 SG S1 SG S1 SG S1 SG S1 SG S1 SG SG SG TTP TTP TTP NG NG NG CAN CAN CAN O2 O2 O2 P2 P2 P2 N2 N2 N2 O3 O3 O3 Deadlinemet Deadlinemet Deadlinemissed! P3 P3 P3 Transformation: S1 is the first slot, m1 and m2 are sent sooner Transformation: P2 is the high priority process on N2 Optimization Example

12. Optimization Strategies • OptimizeSchedule • Synthesizes the communication and assigns priorities to obtain a schedulable application • Based on a greedy approach • Cost function: degree of schedulability • OptimizeBuffers • Synthesizes the communication and assigns priorities to reduce the total buffer size • Based on a hill-climbing heuristic • Cost function: total buffer size • Straightforward solution • Finds a schedulable application • Does not consider the optimization of the design

13. OptimizeSchedule Can We Improve Schedulability? Cost function: degree of schedulability 120 StraightforwardsolutionDoes not perform optimizations 100 80 Average Percentage Deviation [%] 60 OptimizeSchedule? 40 20 0 Simulated AnnealingNear-optimal values for the degree of schedulability 80 160 240 320 400 Number of processes

14. Case study • Vehicle cruise controller • Distributed over theTT and ET clusters OptimizeBuffers Can We Reduce Buffer Sizes? Cost function: total buffer size 10k OptimizeScheduleDoes not optimize thetotal buffer size 8k 6k OptimizeBuffers? Average total buffer size [k] 4k Simulated AnnealingNear-optimal values for the total buffer size 2k 0 80 160 240 320 400 Number of processes

15. Message and Future Work • Future Work • Explore more design problems • Mapping for multi-clusters • How to partition an application in ET and TT activities? Analysis and optimization methods areneeded for the efficient implementation of applications distributed over interconnectedheterogeneous networks.